103 goosed harness_remember returns remembered:false, so session memory
never carried the time anchor. Prepend buildCurrentTimeAgentPrefix on every
H5 /reply and WeChat agent prompt so the model sees Asia/Shanghai clock.
Co-authored-by: Cursor <cursoragent@cursor.com>
Agent replies were reporting UTC time because 0630007-memind never injected
TKMind 当前时间基准 into session memory. Port the time anchor from the
0630004 work only—no miniapp branch changes—and add a 103 rollback script.
Co-authored-by: Cursor <cursoragent@cursor.com>
Codifies the 103 production release workflow used for the
0701bug003 deploy: doc-mandated pre-release tests, build + dry-run,
release-portal-runtime-prod.sh, full post-deploy verification
(health, /auth/login, all 4 goosed ports, log keyword scan, real
API-driven page-create smoke test), and both a preferred rollback
(redeploy previous good commit) and an emergency manual rollback path.
Passwords are taken from env vars (JOHN_PASSWORD/H5_ACCESS_PASSWORD),
matching the existing scripts/fruit-theme-john-e2e.mjs convention —
never hardcoded in the skill file.
Co-Authored-By: Claude <noreply@anthropic.com>
Diagnosed via production session logs (RDS + shared goose PG) that some
users' page-generation attempts drifted: the agent used apps__create_app
(no public URL) then fabricated a placeholder domain when asked for a
link, and separately burned many turns scraping unreachable/anti-bot
search engines. Steer both flows toward the already-working path instead
of banning tools outright:
- apps__create_app is fine for designing/previewing a page, but the
agent must still write_file the result into public/*.html per the
static-page-publish skill and return a link built from the real
public URL template (no invented domains)
- for real-world lookups, load_skill('web') first and prefer
Bing/360 over retrying google.com or hammering anti-bot sites
Also folds in generate_docx guidance for Word-download attachments
(sandbox-fs tool) that landed in the same files during this pass.
Co-Authored-By: Claude <noreply@anthropic.com>
Add a MindSpace public HTML checker for missing companion docx/PDF paths, wire it into runtime packaging and release-portal-runtime-prod.sh, and document the workflow for agents and ops.
Co-authored-by: Cursor <cursoragent@cursor.com>
Bundle memind-portal-tunnel.sh into the runtime artifact and kickstart
cn.tkmind.memind-portal-tunnel on 103 after Portal health checks so 105
nginx can reach 127.0.0.1:19081 following runtime releases.
Co-authored-by: Cursor <cursoragent@cursor.com>
Codifies the 103 production release workflow used for the
0701bug003 deploy: doc-mandated pre-release tests, build + dry-run,
release-portal-runtime-prod.sh, full post-deploy verification
(health, /auth/login, all 4 goosed ports, log keyword scan, real
API-driven page-create smoke test), and both a preferred rollback
(redeploy previous good commit) and an emergency manual rollback path.
Passwords are taken from env vars (JOHN_PASSWORD/H5_ACCESS_PASSWORD),
matching the existing scripts/fruit-theme-john-e2e.mjs convention —
never hardcoded in the skill file.
Co-Authored-By: Claude <noreply@anthropic.com>
Diagnosed via production session logs (RDS + shared goose PG) that some
users' page-generation attempts drifted: the agent used apps__create_app
(no public URL) then fabricated a placeholder domain when asked for a
link, and separately burned many turns scraping unreachable/anti-bot
search engines. Steer both flows toward the already-working path instead
of banning tools outright:
- apps__create_app is fine for designing/previewing a page, but the
agent must still write_file the result into public/*.html per the
static-page-publish skill and return a link built from the real
public URL template (no invented domains)
- for real-world lookups, load_skill('web') first and prefer
Bing/360 over retrying google.com or hammering anti-bot sites
Also folds in generate_docx guidance for Word-download attachments
(sandbox-fs tool) that landed in the same files during this pass.
Co-Authored-By: Claude <noreply@anthropic.com>
103 goosed harness_remember returns remembered:false, so session memory
never carried the time anchor. Prepend buildCurrentTimeAgentPrefix on every
H5 /reply and WeChat agent prompt so the model sees Asia/Shanghai clock.
Co-authored-by: Cursor <cursoragent@cursor.com>
Agent replies were reporting UTC time because 0630007-memind never injected
TKMind 当前时间基准 into session memory. Port the time anchor from the
0630004 work only—no miniapp branch changes—and add a 103 rollback script.
Co-authored-by: Cursor <cursoragent@cursor.com>
john